nVidia included more math units on the chip and removed a handful of limitations constraining sequences of instructions to attain maximum performance out of NV40. While the NV3X architecture relied heavily on nVidia’s run-time compiler in the driver to run shaders at a reasonable speed, the NV40 architecture is inherently more flexible.

nVidia has also added a dedicated video processor on the NV40 that is about as complex as the original GeForce. (好可怕的VPU….) What can the dedicated video processor do? Finally offload CPU intensive coding from your CPU, on to a very fast and expensive GPU. The result? Imagine real-time 1080i MPEG-4 or DivX encoding done on your computer, virtually regardless of your host CPU speed. (所以以後不管轉檔軟體有沒有拍Intel馬屁, Athlon族都可以轉的很快樂??) The other benefit? nVidia says that the video processor will make it to all NV4X chips, so eventually there will be $79 card with this functionality on it. (連低階卡都會有VPU…呵呵…畫質問題應該可以獲得改善吧?)

All this comes at a cost, the NV40 is bigger than Intel’s largest desktop chip (at 220 million transistors) and you can estimate an almost Itanium-like die size. It doesn’t matter much to end users, but don’t plan on overclocking NV40 much. (超頻幅度不高…哇咧) Also, the company needs a smaller manufacturing process to bring NV40 to the mainstream. (0.09, 0.65?? 問題是連CPU廠都沒有呀…)

Performance-wise, the first NV40 based card, the $500 GeForce 6800 Ultra, is fast. At more than 1280 x 1024, it is at least twice as fast as the 9800XT. ATI is about to respond with Radeon X800, but ATI may break sweat this time around.

ps. 打這篇有夠累的, 別指望我翻整篇…